Network Architecture & Scope
A network is best understood as endpoints, links, forwarding systems, protocol boundaries, and services—not as a label based only on distance. This entry builds physical, logical, administrative, and failure-domain views of the same system and follows one request across them.
A useful network model answers five questions: what originates traffic, what carries it, what forwards it, where protocol state terminates, and which failures or policies affect the same set of paths. Labels such as LAN and WAN are shorthand for recurring answers, not definitions derived from cable length alone.
The same deployment has several simultaneous structures. A rack diagram shows physical attachment; an Ethernet or IP diagram shows logical reachability; an ownership diagram shows who may change each component; a failure analysis groups components that can fail together. Confusing those views is a common cause of incorrect redundancy claims.
The irreducible parts #
- An endpoint originates or consumes an application exchange. It may be a laptop, sensor, server process, virtual machine, or container namespace; “host” is a protocol role, not necessarily a physical box.
- A link offers communication between attached termination points. It may be a copper pair, fiber wavelength, radio channel, virtual tunnel, or an abstract adjacency exposed by a lower layer.
- A forwarding system relays traffic between attachments. A bridge forwards link-layer frames; a router forwards network-layer packets; a middlebox may additionally classify or transform traffic.
- A protocol specifies messages exchanged by peer entities. A service is the capability one layer exposes upward. The distinction matters: several protocols can implement a similar service, and one protocol can be used through different interfaces.
- A control system installs or learns forwarding state. The control plane decides; the data plane applies the resulting rule to each unit of traffic.
RFC 8345 models topology using nodes, links, and termination points. That vocabulary is deliberately generic: an IP topology can be layered over Ethernet, which is layered over physical media. A link in the higher model is therefore a service supplied by the lower model, not necessarily one cable.
Scope labels are operational hints #
| Label | Useful interpretation | What it does not guarantee |
|---|---|---|
| PAN | devices organized around one person or very small local system | one protocol, one owner, or short radio range |
| LAN | a local link-layer system under relatively tight administration | one broadcast domain, one building, or absence of routing |
| campus | several local networks joined under a common site design | one failure domain or one security policy |
| WAN | connectivity across sites or provider domains | low speed, public ownership, or a particular switching technology |
| overlay | logical nodes and links built over another network | independence from underlay capacity and failures |
The decisive properties are usually administrative boundary, round-trip delay, capacity, trust, and shared risk. Two racks in one room can be separated by a routed policy boundary; two continents can appear as one private overlay. State those properties directly when they affect a design instead of relying on the scope label.
Follow one application exchange #
Suppose a browser opens a service name:
- The endpoint resolves the name and selects a destination address.
- The operating system chooses a next hop and creates link-layer frames for the local attachment.
- A bridge may relay those frames inside the local IEEE 802 network without terminating the IP exchange.
- A router removes the incoming link framing, forwards the IP packet according to its routing state, and creates new framing for the next link.
- Load balancers, firewalls, or tunnels may introduce additional policy and state boundaries.
- At the destination, transport and application endpoints consume the exchange; the physical server is only one possible hosting boundary.
This walk separates path from session. Link-layer identities and frames can change at every routed hop, while the IP and transport conversation usually spans multiple links. An application name may resolve differently without changing the local access network at all.
Physical, logical, administrative, and failure views #
Draw all four for a production design:
- Physical: ports, radios, media, devices, power, and location.
- Logical: broadcast domains, subnets, tunnels, routes, and forwarding adjacencies.
- Administrative: ownership, configuration authority, trust, and change process.
- Failure: common power, conduit, chassis, software, control plane, and provider dependencies.
Redundant logical paths are not independent when they traverse the same conduit or depend on the same controller. Conversely, one physical fabric can support multiple isolated logical networks. Availability claims belong to the failure view, while reachability claims belong to the logical view.
A compact design checklist #
- Identify the actual protocol endpoints; do not infer them from boxes on a rack diagram.
- Mark every layer transition and the state terminated there.
- Record capacity and propagation constraints on the path, including shared lower-layer resources.
- Draw policy and trust boundaries separately from routing boundaries.
- Name shared-risk groups: power, fiber path, device, software image, controller, and provider.
- Verify both directions. Reachability, filtering, and path capacity may be asymmetric.
The scope label is useful only after these facts are known. Architecture begins with explicit roles and boundaries; LAN, WAN, campus, and overlay then become compact summaries rather than substitutes for analysis.